| 摘要: | 森林是重要的生態系統，提供生物圈豐富的自然資源。森林調查能持續協助政府對
森林發展進行決策，例如：變更、改進和管理。森林的田野調查工作相當繁重且耗時，需
要大量的人力資源和成本，特別是在大範圍的山區。具可信度的森林調查結果能提供人們
準確且即時的資訊，使得森林規劃更有效率、更完整。近年來，遙測技術被應用於大規模
森林調查之應用上，故本研究欲以遙測資料產製森林調查圖，並具有森林樹種的屬性資料。
本研究使用最大熵法(Maximum Entropy Method, MaxEnt)分類庫蘇古爾省(Huwsgul)
額爾德尼布勒甘縣的森林樹種。研究區位於蒙古北部高山地區，森林面積共4,230.1 平方
公里，占額爾德尼布勒甘縣總面積約85%，本研究測試資料為2011 年7 月的Landsat-5 衛
星影像、空間解析度30 公尺的數值高程模型(Digital Elevation Model, DEM)，以及森林組
織公司提供的森林樹種調查圖。本研究使用Landsat-5 衛星影像和數值高程模型，以最大
熵法分類森林樹種和其分布，並進行三項測試：(1)以Landsat 多波段影像分類森林樹種；
(2)以數值高程模型衍生的地理變數分類森林樹種；(3)整合Landsat 多波段影像和地理變數
分類森林樹種。並將測試成果和森林樹種調查圖比較，進行精度評估。
研究成果顯示，僅用多波段影像分類六個樹種的總體精度為69%，Kappa 值為0.35，
僅用地理資料分類六個樹種的總體精度為65%，Kappa 值為0.28，而整合兩種資料後的分
類總體精度為80%，Kappa 值為0.48。根據上述成果比較，整合後的多波段影像資料和地
理資料，能有效提升森林樹種分類的精度。;Forest is a very important ecosystem and natural resource for living things. Based on forest
inventories, government is able to make decisions to converse, improve and manage forests in a
sustainable way. Field work for forestry investigation is difficult and time consuming, because it
needs intensive physical labor and costs, especially surveying in a widely and remotely
mountainous area. A reliable forest inventory can give us more accurate and timely information to
develop new and efficient approaches of forest management. The remote sensing technology has
been recently used for forest investigation for large scale. To produce an informative forest
inventory, forest attributes, including tree species are necessarily investigated.
This research focuses on the classification of forest tree species in Erdenebulgan county,
Huwsgul province, Mongolia, using Maximum entropy method. The study area covers a forest area
of 4230.1 km2 which is almost 85% of total area of Erdenebulgan county and located in a high
mountain region in northern Mongolia. For this study, Landsat 5 satellite imagery taken in July,
2011 and a 30 m DEM (Digital Elevation Model) were acquired to perform image classification.
The forest tree species inventory map collected from Forest Organization Company. Landsat
images and DEM were processed for tree species classification, and a maximum entropy model,
MaxEnt, for predicting the distribution of tree species was applied in this study. This study has
tried three different experiments: (1) spectral bands from Landsat were used for free species
classification; (2) topographical variables were used for tree species classification; and (3) tree
species classification generated from both spectral bands and topographical data. All experimental
results were compared with the tree species inventory to access the mapping accuracy.
The result shows that six different tree species were classified. The overall accuracy from only
spectral bands is 69 % and kappa coefficient is 0.35, and the result from only topographical data
shows 65 % overall accuracy and 0.28 kappa coefficient. Meanwhile, the overall accuracy from
integration of spectral bands and topographical data is 80 % with kappa coefficient of 0.48,
indicating that the integration of topographic data and image data can improve the classification of
tree species in this study area. |
